Dr. Emily Dolson

emilydolson.bsky.social

06 December 2024 09:56

Je voudrais recorder plus videos! Mon francais est mauvaise, mais peut-etre un challenge amusante!

Dr. Emily Dolson

emilydolson.bsky.social

06 December 2024 01:46

That's what we investigated here! We created an environment with three tasks chosen such that we expected Markov Brains to be the best representation for one, Cartesian Genetic Programming to be the best representation for another, and neural networks to not be great at either. 6/n

Dr. Emily Dolson

emilydolson.bsky.social

06 December 2024 01:46

Getting this specialization to happen can be challenging, though. Additionally, based on the earlier results, it's probably not reasonable to assume that all the different sub-problems share an optimal problem solution representation. Can we take advantage of the 2nd problem to solve the 1st? 5/n

Dr. Emily Dolson

emilydolson.bsky.social

06 December 2024 01:46

Max noticed a way to expand this idea to evolutionary swarm robotics, specifically evolution of controllers for heterogenous swarms. A potential benefit of heterogeneous swarms is the possibility of different controllers specializing on different sub-components of the swarm's overall goal. 4/n

Dr. Emily Dolson

emilydolson.bsky.social

06 December 2024 01:46

A few years ago, some folks including @cliffbohm.bsky.social showed that different representations are better for different problems. Moreover, they showed that if you let evolution chose among representations, it will mix and match and choose whichever is best link.springer.com/chapter/10.1... 3/n